Multi fault classification in electrical transmission lines using feature engineering based on autogluon framework

Abstract

With the increasing electricity demand, power transmission lines continue to grow and become increasingly complex. Because even the smallest faults occurring on growing power lines can impact the grid, rapid fault detection, classification, and subsequent repair are crucial. In this study, a new Machine Learning approach based on the Autogluon framework is proposed for rapid fault detection and high-accuracy classification of transmission line faults. Three different methods are employed within the Autogluon framework, and the results are evaluated and compared using ROC analysis. The first method uses the original dataset containing 7861 data points obtained from the open-source Kaggle platform. The second approach adds statistical properties of current and voltage values (mean, standard deviation, and range) to the original dataset. The third method uses the SMOTE algorithm to generate synthetic data and increase the data size to address the imbalance in the number of fault classes in the dataset. The proposed method demonstrates the critical role of data processing and feature engineering in optimizing classification performance and achieving the most accurate diagnosis for multi-label fault classification. The numerical results compared with the literature show that the proposed method is promising for practical applications.